While many people associate artificial intelligence (AI) in the automotive industry with autonomous vehicles, it’s actually a powerful tool that’s driving software development too. I recently joined James Carter and David Fidalgo on the Byte Off Podcast to talk about the impact AI is having across the industry.
AI has the potential to improve the outcomes for quality and support engineers during the automotive software development process. Here at Aurora Labs, we’re using AI to recognize patterns in the behavior of the software, as how it behaves indicates how it runs. By identifying these patterns, you can begin to predict when and how a piece of software might fail before it actually does.
Using the right AI tools (such as Vehicle Software Intelligence) we’re able to help car manufacturers find problems in their vehicles before they cause failures. This allows them to focus on improving quality instead of running around trying to fix problems.
The challenge in using AI tools across these areas is correctly identifying when it’s appropriate. A lot of people see this technology as a silver bullet that will fix all sorts of problems, but it’s actually most powerful in areas where the inputs are unknown or the variables are great.
This is why it’s so often associated with autonomous driving because the technology has to be smart enough to understand that every road, every car, and every tree looks different and still be able to identify them as such. The technology needs to be able to recognize these patterns and learn from the information it is fed.
The shift left
What we’re seeing now is a shift left, which means we’re starting to use AI much earlier in the development process. The idea is to catch problems earlier as this makes them easier to fix, keeps costs down, and saves valuable time. It’s similar to the process of building a house. If you find a problem in the construction of the walls and identify this early on, it’s much cheaper to fix the issue than if the issue had been discovered after the house was complete.
The shift left in the automotive development world is similar. It’s about moving your quality tools and insights earlier in the process so you’re not leaving everything until the end. Fixing issues early on is much less expensive than patching them with over-the-air updates or worse, recalling your vehicles.
- More than one-fifth of industry experts expect software sales to account for at least ten percent of carmakers’ sales as early as 2027
- More than half (53%) of respondents expect setback for electric offensive due to semiconductor shortage
- 45% of respondents expect each connected vehicle to receive up to six OTA updates per year from 2025 onwards
There are other trends influencing this shift. Both the move to CI/CD and agile software development methodology play a role. This means a car that might have been designed over six years, for example, can now be designed in a much shorter period. With these shorter development cycles, it’s vital manufacturers are testing their software early enough in the process so as not to cause delays further down the line.
Another trend is the move toward the software-defined vehicle. With the software disconnected from the hardware platform and any specific model year, there needs to be more focus on the quality of that technology as it’s driving so much within a vehicle – even across different models and generations, in some cases. With this, CI/CD, and agile workflows, there’s an openness to try new AI tools to improve quality and give actionable insights early on at a much lower cost than you might have with more traditional development methods.
Testing the modern vehicle
Because of the complexity of a modern vehicle and now, the option to add features via a subscription, existing testing methods become far more difficult. If you’re trying to write test scenarios for every permutation of variation and in every configuration, you can very quickly get to a point where an engineer physically can’t write all these tests – and you certainly don’t have enough time to run them, even with automation tools.
AI algorithms, however, can monitor the behavior of the software as it’s being run and pick up on deviations automatically. Without any manually defined thresholds, the AI is able to detect changes in behavior. This allows engineers to focus their attention on what is changing and what could potentially affect the vehicle quality and performance.
This benefits both end-users and OEMs. The customer gets their update or subscription feature immediately and can trust that the new software isn’t going to affect something else in the vehicle. Manufacturers, on the other hand, are able to improve quality quickly and more affordably while keeping customer satisfaction high.
Artificial intelligence is a powerful tool and something the industry is becoming increasingly open to. If you’d like to find out more about automotive software quality assurance take a look here.